Application of artificial intelligence to predict compound toxicity and challenge to federated learning in drug discovery

Abstract

In this talk, I will present the first attempt to apply Explainable AI techniques to visualize what AI learns from 2D chemical structure images, which demonstrated that AI sees chemically meaningful substructures in the image to make a prediction as chemists do. In addition, I will talk about recent progress and future prospect of Federated Learning (FL). Astellas is participating in the MELLODDY consortium, aiming to develop privacy-preserved FL platform to leverage each company's data for building predictive models for drug discovery while owners keep control of their proprietary data.